Internal combustion (IC) engines still power most of the vehicles on road and will likely to remain so in the near future, especially for heavy duty applications in which electrification is typically more challenging. Therefore, continued improvements on IC engines in terms of efficiency and longevity are necessary for a more sustainable transportation sector. Two important design objectives for heavy duty engines with wet liners are to reduce friction loss and to lower the risks of cavitation damages, both of which can be greatly influenced by the piston-liner clearance and the design of the piston skirt. However, engine design optimization is difficult due to the nonlinear interactions between the key design variables and the design objectives, as well as the multi-physics and multi-scale nature of the mechanisms that are relevant to the design objectives. In this work, an efficient optimization method is proposed to optimize the piston skirt design and the piston-liner warm clearance with respect to the aforementioned objectives, subject to design constraints. The method couples a neural network surrogate of a high-fidelity numerical model for piston secondary motion, which can accurately predict the objective values corresponding to different piston skirt designs and piston-liner clearances, with genetic algorithms to seek designs that can achieve a good tradeoff between the two objectives, while having a reasonable computation cost. Results show that with the proposed method, the optimal skirt design can have up to 62% lower risk of cavitation damage and up to 42% lower friction loss compared to the baseline design while satisfying important design constraints.